Analiza metoda generiranja sintetičkih skupova podataka

undergraduate thesis

• Title (english): Survey of methods for synthetic dataset generation
• Author: Dominik Sekula
• Mentor: Matko Orsag (mentor)
• Committee member: Matko Orsag (predsjednik povjerenstva)
• Committee member: Marsela Polić (član povjerenstva)
• Committee member: Jelena Vuletić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-07-15, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Cilj rada je navesti, provesti i analizirati metode generacije sintetičkih skupova podataka. Rad se prvotno bavi analizom i prikazivanjem sučelja za generaciju i upravljanje skupovima podataka pod nazivom NDDS (Nvidia Deep learning Dataset Synthesizer) korištenim u svrhu stvaranju podataka za učenje dubokih konvolucijskih neuronskih mreža. Mreže su trenirane na dva načina: samo na sintetičkim podacima i uz korištenje realnih podataka za validaciju. Uspoređivanje uspješnosti se baziralo na točnosti tj. postotku preklapanja između ciljne maske i dobivene predikcije. Postupak je pokazao prihvatljivu točnost na dijelu podataka te ukazao na važnost provjere razičitih metoda treniranja.

Abstract (english)

The aim of this thesis is to list, implement and analyze the methods for synthetic dataset generation. Interface for the generation and management of datasets called NDDS (Nvidia Deep learning Dataset Synthesizer) is analyzed and used for the purpose of creating data for learning deep convolutional neural networks. Networks are trained in two ways: on synthetic data only and on synthetic data with the addition of the real images used for the validation purpose. The comparison of performance was based on IoU score. Namely, the percentage of overlap between the target mask and the obtained prediction mask was calculated for each model. The procedure showed acceptable accuracy on some parts of the data and pointed out the importance of testing different training methods

• Keywords: semantička segmentacija; duboko učenje; sintetički skup podataka; NDDS
• Keywords (english): semantic segmentation; deep learning; synthetic dataset; NDDS
• Language: croatian
• URN:NBN: urn:nbn:hr:168:910947
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-15 11:11:58